On Wed, Nov 7, 2018 at 4:49 PM J. Bruce Fields <bfields@xxxxxxxxxxxx> wrote:
>
> On Fri, Oct 19, 2018 at 11:29:05AM -0400, Olga Kornievskaia wrote:
> > From: Olga Kornievskaia <kolga@xxxxxxxxxx>
> >
> > Given a universal address, mount the source server from the destination
> > server. Use an internal mount. Call the NFS client nfs42_ssc_open to
> > obtain the NFS struct file suitable for nfsd_copy_range.
> >
> > Ability to do "inter" server-to-server depends on the an nfsd kernel
> > parameter "inter_copy_offload_enabled".
> >
> > Signed-off-by: Andy Adamson <andros@xxxxxxxxxx>
> > Signed-off-by: Olga Kornievskaia <kolga@xxxxxxxxxx>
> > ---
> > fs/nfsd/nfs4proc.c | 298 ++++++++++++++++++++++++++++++++++++++++++++++++---
> > fs/nfsd/nfssvc.c | 6 ++
> > fs/nfsd/xdr4.h | 5 +
> > include/linux/nfs4.h | 1 +
> > 4 files changed, 293 insertions(+), 17 deletions(-)
> >
> > diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
> > index 59e9d0c..6dcd80c 100644
> > --- a/fs/nfsd/nfs4proc.c
> > +++ b/fs/nfsd/nfs4proc.c
> > @@ -1153,6 +1153,229 @@ void nfsd4_shutdown_copy(struct nfs4_client *clp)
> > while ((copy = nfsd4_get_copy(clp)) != NULL)
> > nfsd4_stop_copy(copy);
> > }
> > +#ifdef CONFIG_NFSD_V4_2_INTER_SSC
> > +
> > +extern struct file *nfs42_ssc_open(struct vfsmount *ss_mnt,
> > + struct nfs_fh *src_fh,
> > + nfs4_stateid *stateid);
> > +extern void nfs42_ssc_close(struct file *filep);
> > +
> > +extern void nfs_sb_deactive(struct super_block *sb);
> > +
> > +#define NFSD42_INTERSSC_MOUNTOPS "minorversion=2,vers=4,addr=%s,clientaddr=%s"
>
> The nfs man page says "clientaddr=" has no effect on 4.2 mounts.
I only have nfs man page from RHEL7.5 and I don't see that.
> Also, what's the "addr=" option for, isn't the server address already
> given in the mount string? (Honest question, I may be wrong here.)
I believe going thru the kernel vfs_kern_mount() we need to specify
"addr=" otherwise it doesn't know which server to mount.
>
> > +
> > +/**
> > + * Support one copy source server for now.
> > + */
> > +static struct vfsmount *
> > +nfsd4_interssc_connect(struct nl4_server *nss, struct svc_rqst *rqstp)
> > +{
> > + struct file_system_type *type;
> > + struct vfsmount *ss_mnt;
> > + struct nfs42_netaddr *naddr;
> > + struct sockaddr_storage tmp_addr;
> > + size_t tmp_addrlen, match_netid_len = 3;
> > + char *startsep = "", *endsep = "", *match_netid = "tcp";
> > + char *ipaddr, *ipaddr2, *raw_data;
> > + int len, raw_len, status = -EINVAL;
> > +
> > + /* Currently support only NL4_NETADDR source server */
> > + if (nss->nl4_type != NL4_NETADDR) {
> > + WARN(nss->nl4_type != NL4_NETADDR,
> > + "nfsd4_copy src server not NL4_NETADDR\n");
>
> Won't nfsd4_decode_nl4_server actually let through NL4_NAME and NL4_URL?
Yes. I think the logic would be not to limit the xdr functionality
from not parsing it as if the support in the main code the xdr code
doesn't change.
> That would make this WARN() triggerable by a client--that's bad.
Why? Would you rather it silently failed?
> > + goto out_err;
> > + }
> > +
> > + naddr = &nss->u.nl4_addr;
> > +
> > + tmp_addrlen = rpc_uaddr2sockaddr(SVC_NET(rqstp), naddr->addr,
> > + naddr->addr_len,
> > + (struct sockaddr *)&tmp_addr,
> > + sizeof(tmp_addr));
> > + if (tmp_addrlen == 0)
> > + goto out_err;
> > +
> > + if (tmp_addr.ss_family == AF_INET6) {
> > + startsep = "[";
> > + endsep = "]";
> > + match_netid = "tcp6";
> > + match_netid_len = 4;
> > + }
> > +
> > + if (naddr->netid_len != match_netid_len ||
> > + strncmp(naddr->netid, match_netid, naddr->netid_len))
>
> Just strcmp(naddr->netid, match_netid) would do the job.
Will change.
> > + goto out_err;
> > +
> > + /* Construct the raw data for the vfs_kern_mount call */
> > + len = RPC_MAX_ADDRBUFLEN + 1;
> > + ipaddr = kzalloc(len, GFP_KERNEL);
> > + if (!ipaddr)
> > + goto out_err;
> > +
> > + rpc_ntop((struct sockaddr *)&tmp_addr, ipaddr, len);
> > +
> > + /* 2 for ipv6 endsep and startsep. 3 for ":/" and trailing '/0'*/
> > + ipaddr2 = kzalloc(len + 5, GFP_KERNEL);
> > + if (!ipaddr2)
> > + goto out_free_ipaddr;
> > +
> > + rpc_ntop((struct sockaddr *)&rqstp->rq_daddr, ipaddr2, len + 5);
>
> Replace the above by two calls to a function that does kmalloc+rpcntop?
> (Though actually I don't think we need ipaddr.)
Will do.
> > +
> > + raw_len = strlen(NFSD42_INTERSSC_MOUNTOPS) + strlen(ipaddr) +
> > + strlen(ipaddr2);
> > + raw_data = kzalloc(raw_len, GFP_KERNEL);
> > + if (!raw_data)
> > + goto out_free_ipaddr2;
> > +
> > + snprintf(raw_data, raw_len, NFSD42_INTERSSC_MOUNTOPS, ipaddr,
> > + ipaddr2);
> > +
> > + status = -ENODEV;
> > + type = get_fs_type("nfs");
> > + if (!type)
> > + goto out_free_rawdata;
>
> I believe you also need a put_filesystem after this. (e.g. see
> kernel/trace/trace.c:trace_automount().)
Got it. Thanks.
>
> > +
> > + /* Set the server:<export> for the vfs_kerne_mount call */
> > + memset(ipaddr2, 0, len + 5);
> > + snprintf(ipaddr2, len + 5, "%s%s%s:/", startsep, ipaddr, endsep);
> > +
> > + dprintk("%s Raw mount data: %s server:export %s\n", __func__,
> > + raw_data, ipaddr2);
> > +
> > + /* Use an 'internal' mount: MS_KERNMOUNT -> MNT_INTERNAL */
> > + ss_mnt = vfs_kern_mount(type, MS_KERNMOUNT, ipaddr2, raw_data);
> > + if (IS_ERR(ss_mnt)) {
> > + status = PTR_ERR(ss_mnt);
> > + goto out_free_rawdata;
> > + }
> > +
>
> Let's combine the successful and failure cases, so the below should be
> something like:
>
> out_free_rawdata:
> kfree(raw_data);
> out_free_ipaddr2:
> kfree(ipaddr2);
> out_free_ipaddr:
> kfree(ipaddr);
> out_err:
> if (IS_ERR(ret))
> dprintk("--> %s ERROR %d\n", __func__, status);
> return ret;
Ok will do.
>
>
> > + kfree(raw_data);
> > + kfree(ipaddr2);
> > + kfree(ipaddr);
> > +
> > + return ss_mnt;
> > +
> > +out_free_rawdata:
> > + kfree(raw_data);
> > +out_free_ipaddr2:
> > + kfree(ipaddr2);
> > +out_free_ipaddr:
> > + kfree(ipaddr);
> > +out_err:
> > + dprintk("--> %s ERROR %d\n", __func__, status);
> > + return ERR_PTR(status);
> > +}
> > +
> > +static void
> > +nfsd4_interssc_disconnect(struct vfsmount *ss_mnt)
> > +{
> > + nfs_sb_deactive(ss_mnt->mnt_sb);
> > + mntput(ss_mnt);
> > +}
> > +
> > +/**
> > + * nfsd4_setup_inter_ssc
> > + *
> > + * Verify COPY destination stateid.
> > + * Connect to the source server with NFSv4.1.
> > + * Create the source struct file for nfsd_copy_range.
> > + * Called with COPY cstate:
> > + * SAVED_FH: source filehandle
> > + * CURRENT_FH: destination filehandle
> > + *
> > + * Returns errno (not nfserrxxx)
> > + */
> > +static struct vfsmount *
> > +nfsd4_setup_inter_ssc(struct svc_rqst *rqstp,
> > + struct nfsd4_compound_state *cstate,
> > + struct nfsd4_copy *copy)
> > +{
> > + struct svc_fh *s_fh = NULL;
> > + stateid_t *s_stid = ©->cp_src_stateid;
> > + struct vfsmount *ss_mnt;
> > + __be32 status;
> > +
> > + /* Verify the destination stateid and set dst struct file*/
> > + status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
> > + ©->cp_dst_stateid,
> > + WR_STATE, ©->file_dst, NULL,
> > + NULL);
> > + if (status) {
> > + ss_mnt = ERR_PTR(be32_to_cpu(status));
>
> That looks wrong. I don't think IS_ERR() is going to be true for that
> value.
>
> If we need to return either an nfserr or a pointer, best is probably to
> have the function return __be32 and have the pointer returned in an
> argument.
>
> (Thought I notice the only caller ignores the error value, I wonder if
> that's right.)
>
> > + goto out;
> > + }
> > +
> > + ss_mnt = nfsd4_interssc_connect(copy->cp_src, rqstp);
> > + if (IS_ERR(ss_mnt))
> > + goto out;
>
> So this function can return -ERRNO, or nfserr_*, or a pointer? That
> won't work.
Ok I'll change the function to return the __be32 always. And return
pointer will be one of the args.
>
> > + s_fh = &cstate->save_fh;
> > +
> > + copy->c_fh.size = s_fh->fh_handle.fh_size;
> > + memcpy(copy->c_fh.data, &s_fh->fh_handle.fh_base, copy->c_fh.size);
> > + copy->stateid.seqid = s_stid->si_generation;
> > + memcpy(copy->stateid.other, (void *)&s_stid->si_opaque,
> > + sizeof(stateid_opaque_t));
> > +
> > +out:
> > + return ss_mnt;
> > +}
> > +
> > +static void
> > +nfsd4_cleanup_inter_ssc(struct vfsmount *ss_mnt, struct file *src,
> > + struct file *dst)
> > +{
> > + nfs42_ssc_close(src);
> > + fput(src);
> > + fput(dst);
> > + mntput(ss_mnt);
> > +}
> > +
> > +#else /* CONFIG_NFSD_V4_2_INTER_SSC */
> > +
> > +static struct vfsmount *
> > +nfsd4_setup_inter_ssc(struct svc_rqst *rqstp,
> > + struct nfsd4_compound_state *cstate,
> > + struct nfsd4_copy *copy)
> > +{
> > + return ERR_PTR(-EINVAL);
>
> I wonder if that's really the right error for the
> server-to-server-copy-unsupported case.
Should be not_supported because COPY itself is not supported. If COPY
was supported but failed for whatever reason we couldn't mount then
the error should be OFFLOAD_DENIED.
>
> > +}
> > +
> > +static void
> > +nfsd4_cleanup_inter_ssc(struct vfsmount *ss_mnt, struct file *src,
> > + struct file *dst)
> > +{
> > +}
> > +
> > +static void
> > +nfsd4_interssc_disconnect(struct vfsmount *ss_mnt)
> > +{
> > +}
> > +
> > +static struct file *nfs42_ssc_open(struct vfsmount *ss_mnt,
> > + struct nfs_fh *src_fh,
> > + nfs4_stateid *stateid)
> > +{
> > + return NULL;
> > +}
> > +#endif /* CONFIG_NFSD_V4_2_INTER_SSC */
> > +
> > +static __be32
> > +nfsd4_setup_intra_ssc(struct svc_rqst *rqstp,
> > + struct nfsd4_compound_state *cstate,
> > + struct nfsd4_copy *copy)
> > +{
> > + return nfsd4_verify_copy(rqstp, cstate, ©->cp_src_stateid,
> > + ©->file_src, ©->cp_dst_stateid,
> > + ©->file_dst, NULL);
> > +}
> > +
> > +static void
> > +nfsd4_cleanup_intra_ssc(struct file *src, struct file *dst)
> > +{
> > + fput(src);
> > + fput(dst);
> > +}
> >
> > static void nfsd4_cb_offload_release(struct nfsd4_callback *cb)
> > {
> > @@ -1217,12 +1440,16 @@ static __be32 nfsd4_do_copy(struct nfsd4_copy *copy, bool sync)
> > status = nfs_ok;
> > }
> >
> > - fput(copy->file_src);
> > - fput(copy->file_dst);
> > + if (copy->cp_src) /* Inter server SSC */
> > + nfsd4_cleanup_inter_ssc(copy->ss_mnt, copy->file_src,
> > + copy->file_dst);
> > + else
> > + nfsd4_cleanup_intra_ssc(copy->file_src, copy->file_dst);
> > +
> > return status;
> > }
> >
> > -static void dup_copy_fields(struct nfsd4_copy *src, struct nfsd4_copy *dst)
> > +static int dup_copy_fields(struct nfsd4_copy *src, struct nfsd4_copy *dst)
> > {
> > dst->cp_src_pos = src->cp_src_pos;
> > dst->cp_dst_pos = src->cp_dst_pos;
> > @@ -1232,8 +1459,21 @@ static void dup_copy_fields(struct nfsd4_copy *src, struct nfsd4_copy *dst)
> > memcpy(&dst->fh, &src->fh, sizeof(src->fh));
> > dst->cp_clp = src->cp_clp;
> > dst->file_dst = get_file(src->file_dst);
> > - dst->file_src = get_file(src->file_src);
> > + if (!src->cp_src) /* for inter, file_src doesnt exist yet */
> > + dst->file_src = get_file(src->file_src);
> > memcpy(&dst->cp_stateid, &src->cp_stateid, sizeof(src->cp_stateid));
> > + if (src->cp_src) {
> > + dst->cp_src = kmalloc(sizeof(struct nl4_server), GFP_KERNEL);
> > + if (!dst->cp_src)
> > + return -ENOMEM;
> > + memcpy(dst->cp_src, src->cp_src, sizeof(struct nl4_server));
> > + }
> > + memcpy(&dst->stateid, &src->stateid, sizeof(src->stateid));
> > + memcpy(&dst->c_fh, &src->c_fh, sizeof(src->c_fh));
> > + dst->ss_mnt = src->ss_mnt;
> > +
> > + return 0;
> > +
> > }
> >
> > static void cleanup_async_copy(struct nfsd4_copy *copy)
> > @@ -1244,6 +1484,7 @@ static void cleanup_async_copy(struct nfsd4_copy *copy)
> > spin_lock(©->cp_clp->async_lock);
> > list_del(©->copies);
> > spin_unlock(©->cp_clp->async_lock);
> > + kfree(copy->cp_src);
> > nfs4_put_copy(copy);
> > }
> >
> > @@ -1252,7 +1493,18 @@ static int nfsd4_do_async_copy(void *data)
> > struct nfsd4_copy *copy = (struct nfsd4_copy *)data;
> > struct nfsd4_copy *cb_copy;
> >
> > + if (copy->cp_src) { /* Inter server SSC */
> > + copy->file_src = nfs42_ssc_open(copy->ss_mnt, ©->c_fh,
> > + ©->stateid);
> > + if (IS_ERR(copy->file_src)) {
> > + copy->nfserr = nfserr_offload_denied;
> > + nfsd4_interssc_disconnect(copy->ss_mnt);
> > + goto do_callback;
> > + }
> > + }
> > +
> > copy->nfserr = nfsd4_do_copy(copy, 0);
> > +do_callback:
> > cb_copy = kzalloc(sizeof(struct nfsd4_copy), GFP_KERNEL);
> > if (!cb_copy)
> > goto out;
> > @@ -1276,11 +1528,19 @@ static int nfsd4_do_async_copy(void *data)
> > __be32 status;
> > struct nfsd4_copy *async_copy = NULL;
> >
> > - status = nfsd4_verify_copy(rqstp, cstate, ©->cp_src_stateid,
> > - ©->file_src, ©->cp_dst_stateid,
> > - ©->file_dst, NULL);
> > - if (status)
> > - goto out;
> > + if (copy->cp_src) { /* Inter server SSC */
> > + if (!inter_copy_offload_enable || copy->cp_synchronous) {
> > + status = nfserr_notsupp;
> > + goto out;
> > + }
> > + copy->ss_mnt = nfsd4_setup_inter_ssc(rqstp, cstate, copy);
> > + if (IS_ERR(copy->ss_mnt))
> > + return nfserr_offload_denied;
>
> We should check that this is the right error to return in all those
> failure cases.
Well once I change nfsd4_setup_inter_ssc() to return an error itself
it'll just return status. But I'll double check the error returns.
> That's all I have for now.
Thank you for the reviews. I'm working on the next version. But in
addition to this, I need the VFS piece with this patch series now
because server piece needs the generic cross filesystem
copy_file_range() support via do_splice because the server reads out
of NFS and writes into the local file system.
>
> --b.
>
> > + } else {
> > + status = nfsd4_setup_intra_ssc(rqstp, cstate, copy);
> > + if (status)
> > + return status;
> > + }
> >
> > copy->cp_clp = cstate->clp;
> > memcpy(©->fh, &cstate->current_fh.fh_handle,
> > @@ -1291,15 +1551,15 @@ static int nfsd4_do_async_copy(void *data)
> > status = nfserrno(-ENOMEM);
> > async_copy = kzalloc(sizeof(struct nfsd4_copy), GFP_KERNEL);
> > if (!async_copy)
> > - goto out;
> > - if (!nfs4_init_cp_state(nn, copy)) {
> > - kfree(async_copy);
> > - goto out;
> > - }
> > + goto out_err;
> > + if (!nfs4_init_cp_state(nn, copy))
> > + goto out_err;
> > refcount_set(&async_copy->refcount, 1);
> > memcpy(©->cp_res.cb_stateid, ©->cp_stateid,
> > sizeof(copy->cp_stateid));
> > - dup_copy_fields(copy, async_copy);
> > + status = dup_copy_fields(copy, async_copy);
> > + if (status)
> > + goto out_err;
> > async_copy->copy_task = kthread_create(nfsd4_do_async_copy,
> > async_copy, "%s", "copy thread");
> > if (IS_ERR(async_copy->copy_task))
> > @@ -1310,13 +1570,17 @@ static int nfsd4_do_async_copy(void *data)
> > spin_unlock(&async_copy->cp_clp->async_lock);
> > wake_up_process(async_copy->copy_task);
> > status = nfs_ok;
> > - } else
> > + } else {
> > status = nfsd4_do_copy(copy, 1);
> > + }
> > out:
> > return status;
> > out_err:
> > cleanup_async_copy(async_copy);
> > - goto out;
> > + status = nfserrno(-ENOMEM);
> > + if (copy->cp_src)
> > + nfsd4_interssc_disconnect(copy->ss_mnt);
> > + goto out_err;
> > }
> >
> > struct nfsd4_copy *
> > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > index 89cb484..9d254e7 100644
> > --- a/fs/nfsd/nfssvc.c
> > +++ b/fs/nfsd/nfssvc.c
> > @@ -30,6 +30,12 @@
> >
> > #define NFSDDBG_FACILITY NFSDDBG_SVC
> >
> > +bool inter_copy_offload_enable;
> > +EXPORT_SYMBOL_GPL(inter_copy_offload_enable);
> > +module_param(inter_copy_offload_enable, bool, 0644);
> > +MODULE_PARM_DESC(inter_copy_offload_enable,
> > + "Enable inter server to server copy offload. Default: false");
> > +
> > extern struct svc_program nfsd_program;
> > static int nfsd(void *vrqstp);
> >
> > diff --git a/fs/nfsd/xdr4.h b/fs/nfsd/xdr4.h
> > index c98ef64..c7e3df1 100644
> > --- a/fs/nfsd/xdr4.h
> > +++ b/fs/nfsd/xdr4.h
> > @@ -546,7 +546,12 @@ struct nfsd4_copy {
> > struct task_struct *copy_task;
> > refcount_t refcount;
> > bool stopped;
> > +
> > + struct vfsmount *ss_mnt;
> > + struct nfs_fh c_fh;
> > + nfs4_stateid stateid;
> > };
> > +extern bool inter_copy_offload_enable;
> >
> > struct nfsd4_seek {
> > /* request */
> > diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
> > index 4d76f87..e53a261 100644
> > --- a/include/linux/nfs4.h
> > +++ b/include/linux/nfs4.h
> > @@ -17,6 +17,7 @@
> > #include <linux/uidgid.h>
> > #include <uapi/linux/nfs4.h>
> > #include <linux/sunrpc/msg_prot.h>
> > +#include <linux/nfs.h>
> >
> > enum nfs4_acl_whotype {
> > NFS4_ACL_WHO_NAMED = 0,
> > --
> > 1.8.3.1
